add static obstacle support to the aviary

[evanofficial]

Re #60. Scoped to static obstacles for this round — the issue mentions lidar/depth as a natural follow-up but those felt like a much bigger conversation about the action/observation space, so this PR keeps things minimal and additive. Happy to take a swing at sensors in a follow-up once this lands and the API shape feels right.

What this adds

• PyFlyt.core.abstractions.Obstacle — thin wrapper around loadURDF with useFixedBase=True, takes a position/orientation/scale and exposes the PyBullet body Id.
• Aviary.add_obstacle(urdf, position, orientation=None, scale=1.0) — registers the obstacle with the aviary, calls register_all_new_bodies() so collisions get tracked, and remembers the spec so it gets respawned on reset().
• Three built-in primitive URDFs under PyFlyt/models/obstacles/: cube, cylinder, sphere. Loadable by name; arbitrary URDF paths also work.

Usage

env = Aviary(start_pos=..., start_orn=..., drone_type="quadx")
env.add_obstacle("cube", position=np.array([2.0, 0.0, 0.5]))
env.add_obstacle("/path/to/wall.urdf", position=np.array([4.0, 0.0, 1.0]))

Collisions show up in env.contact_array the same way drone-drone collisions do.

Tests

tests/test_obstacles.py covers:

• spawning each built-in primitive
• pose round-tripping through PyBullet
• static-under-gravity behaviour
• loading a custom URDF path
• bad-name and bad-shape error paths
• obstacles persisting across Aviary.reset()
• drone↔obstacle collision actually populating contact_array
• Obstacle.builtin_obstacles() listing

All 11 obstacle tests pass locally on Python 3.12 / PyBullet 3.2.7. The existing tests/test_gym_envs.py suite (136 tests) still passes — no regressions.

Docs

Added docs_source/documentation/core/abstractions/obstacle.md and hooked it into the abstractions toctree, matching the existing pattern (autoclass + a usage example).

Follow-ups (not in this PR)

• Dynamic / movable obstacles (mass > 0, no fixed base, optional scripted trajectories)
• Range / depth sensors with obstacle-aware observation spaces

Open to any naming / API tweaks before this gets cleaned up.